Use M_PI to replace 3.14 constants (#16421)

### Description

Use M_PI to replace 3.14 constants

### Motivation and Context

Fixes #16413

onnxruntime/contrib_ops/cuda/diffusion/bias_split_gelu_impl.cu

@@ -41,7 +41,7 @@ __global__ void biasSplitGeluKernel(T const* input, T const* bias, T* output) {
     auto value_right = (float)(input[index_input + HHS] + bias[index_bias + HHS]);
 
     // Gelu is applied to right side only: Gelu(x) = x * 0.5 * (erf(x / sqrt(2)) + 1.0)
-    float gelu_right = value_right * 0.5f * (erff(value_right / 1.41421356237f) + 1.0f);
+    float gelu_right = value_right * 0.5f * (erff(value_right / static_cast<float>(M_SQRT2)) + 1.0f);
     float result = value_left * gelu_right;
     output[index_output] = static_cast<T>(result);
     index_input += TPB;

onnxruntime/core/providers/cpu/ml/ml_common.h

@@ -178,7 +178,7 @@ static inline float ErfInv(float x) {
   float sgn = x < 0 ? -1.0f : 1.0f;
   x = (1 - x) * (1 + x);
   float log = std::log(x);
-  float v = 2 / (3.14159f * 0.147f) + 0.5f * log;
+  float v = 2 / (static_cast<float>(M_PI) * 0.147f) + 0.5f * log;
   float v2 = 1 / (0.147f) * log;
   float v3 = -v + std::sqrt(v * v - v2);
   x = sgn * std::sqrt(v3);
@@ -244,7 +244,7 @@ static inline void multiclass_probability(int64_t classcount,
   }
 }
 
-static constexpr float ml_sqrt2 = 1.41421356f;
+static constexpr float ml_sqrt2 = static_cast<float>(M_SQRT2);
 
 static inline float ComputeLogistic(float val) {
   float v = 1 / (1 + std::exp(-std::abs(val)));

onnxruntime/core/providers/cpu/signal/dft.cc

@@ -79,7 +79,7 @@ static inline T bit_reverse(T num, unsigned significant_bits) {
 template <typename T>
 static T compute_angular_velocity(size_t number_of_samples, bool inverse) {
   // Calculate fundamental angular velocity
-  static constexpr T pi = static_cast<T>(3.14159265);
+  static constexpr T pi = static_cast<T>(M_PI);
   static constexpr T tau = 2 * pi;
   T inverse_switch = inverse ? 1.f : -1.f;
   T angular_velocity = inverse_switch * tau / number_of_samples;
@@ -196,7 +196,7 @@ static Status dft_bluestein_z_chirp(
     OpKernelContext* ctx, const Tensor* X, Tensor* Y, Tensor& b_fft, Tensor& chirp, size_t X_offset, size_t X_stride, size_t Y_offset, size_t Y_stride,
     int64_t axis, size_t dft_length, const Tensor* window, bool inverse, InlinedVector<std::complex<T>>& V,
     InlinedVector<std::complex<T>>& temp_output) {
-  static constexpr T pi = static_cast<T>(3.14159265);
+  static constexpr T pi = static_cast<T>(M_PI);
 
   AllocatorPtr alloc;
   ORT_RETURN_IF_ERROR(ctx->GetTempSpaceAllocator(&alloc));

onnxruntime/core/providers/cpu/signal/window_functions.cc

@@ -53,7 +53,7 @@ struct CosineSumWindow {
     auto* Y_data = reinterpret_cast<T*>(Y->MutableDataRaw());
 
     // Calculate the radians to increment per sample
-    constexpr double pi = 3.14159265;
+    constexpr double pi = M_PI;
     constexpr double tau = 2 * pi;
     const size_t denominator = is_periodic ? size : size - 1;
     const double angular_increment = tau / denominator;

onnxruntime/test/contrib_ops/bias_split_gelu_op_test.cc

@@ -17,7 +17,7 @@ std::vector<float> ComputeGelu(const std::vector<float>& input_data) {
 
   for (size_t i = 0; i < input_data.size(); i++) {
     float x = input_data[i];
-    float y = x * (0.5f * (1.0f + std::erff(x / 1.41421356237f)));
+    float y = x * (0.5f * (1.0f + std::erff(x / static_cast<float>(M_SQRT2))));
     output.push_back(y);
   }
   return output;

winml/test/api/LearningModelSessionAPITest.cpp

@@ -722,7 +722,7 @@ static void DiscreteFourierTransform(
 template <typename T>
 static auto MakePureFrequency(float frequency_in_hertz, size_t signal_size, size_t sample_rate) {
   float amplitude = 4;
-  float angular_velocity = frequency_in_hertz * 2 * 3.1415f;
+  float angular_velocity = frequency_in_hertz * 2 * static_cast<float>(M_PI);
   std::vector<T> signal(signal_size);
   for (size_t i = 0; i < signal_size; i++) {
     T time = i / static_cast<T>(sample_rate);